See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221503838

Position control of a pneumatic levitation system.

Conference Paper · January 2005

DOI: 10.1109/ETFA.2005.1612568 · Source: DBLP

CITATIONS READS

23 2,448

3 authors, including:

Juan Manuel Escaño Francisco R. Rubio

Universidad de Sevilla Universidad de Sevilla
92 PUBLICATIONS 390 CITATIONS 210 PUBLICATIONS 4,519 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Juan Manuel Escaño on 13 March 2014.

The user has requested enhancement of the downloaded file.

 Position Control of a Pneumatic Levitation System

Juan M. Escafio*, Manuel G. Ortega, and Francisco R. Rubio t

Abstract an Ethernet using the MODBUS protocol based on TCP-

The position control of a pneumatic levitation system
is described in this paper The control system uses vi-
sualfeedback to compute the object position, which can be
modified by means of an air blower acted by a variable-
speed drive. The system is employed as a test bed of con-
trol strategies. A performance comparison attained by a
classical PID, a robust Ho, control law and a fuzzy con-
troller is carried out, including tracking and disturbance
rejection experiments.
IP. Finally, the ball position is computed by means of a
camera connected to the PC, which has an appropriate im-
age processing software available [1]. Figure 1 shows a
scheme of the application and a picture of the system is
exposed in figure 2.
position

Pneumatic levitation, robust control, fuzzy control,

control applications.

1 Introduction
The use of pneumatic levitation systems is a not suf- Figure 1. Scheme of the application
ficiently exploited application, perhaps because of their
difficulty to regulate the vertical movements of the float-
ing objects. Nowadays there exist prototypes of this kind
of systems applied, among others, to food carriage (for
example, see [8]). The main advantages of this kind of
transport are the low friction and the material management
without any contact with mechanical parts. Therefore, a
too much cleaner and faster conveyance is achieved with
this kind of devices.
In this work a lab plant, which consists of industrial
components is employed to test several control strategies
applied to pneumatic levitation systems. The main goal is
to fulfill a real time control on the position of an object (in
this case a ball) hovering in a air current, which catches
the ball with some force.
The device is located at the Process Control Labora-
tory of the Altair Technological Center in Seville (Spain),
which consists of a centrifugal fan blower acted by an
AC motor connected to a variable-speed drive. The drive Figure 2. Picture of the system
speed-reference is sent from a trading programmable logic
controller (PLC), which receives the ball position mea-
The remainder of the paper is organized as follows: in
surement from a personal computer linked to it through
Section 2 a brief theoretical study for system modelling
*J.M. Escanio is with the Depto. Electricidad y Automatica, Cen- is presented, and the results of the system identification
tro Tecnol6gico Altair, Barbero de Sevilla,J 41006 Sevilla, (Spain),
- at several operating points are exposed. The description
email:jmesca@e-altair.org of the synthesis of the different controllers, as well as a
tM.G. Ortega and F.R. Rubio are with the Depto. Ingenierfa de Sis-
temas y Automatica, Escuela Tecnica Superior de Ingenieros, Univer-
comparison among the performances achieved, are shown
sidad de Sevilla, Camino de los Descubrimientos, s/n 41092 Sevilla
-
in Section 3. Finally, major conclusions are summarized
(Spain) email: {mortega,rubio} @esi.us.es in Section 4.

0-7803-9402-X/05/$20.00 C 2005 IEEE 523 VOLUME 1

2 System model
The system behavior can be understood by means of
the motion model of a body sunken within the pale of a
gaseous stream. Starting from figure 3, a null gas veloc-
ity at points A and B can be concluded. Besides, veloc-
ity at points C and D is greater than the one at points far
from the ball surface. From Bernoulli's equation, it can
be stated that high air velocity zones are low pressure and
vice versa. Therefore, a pressure difference is produced
from the airflow center to the far away zones, which makes
the ball almost to maintain its lateral position along the air
stream.
F drag AF
T~~~~ Fg
Figure 3. Forces acting on the ball
A complete model of the system may result quite diffi-
cult to obtain. In a simplified manner, it can be stated that
the main forces acting on the ball (see figure 3) are:
* Dragging forces of air resistance: Fdrag
C2D Pa7FT2 V2, where v is the air flow velocity, Pa iS
the air density, r is the ball radius, and CD is the
drag coefficient of the object.
* Forces due to gradients of air pressure: Fvp
2
3 3r Pav
P dh' where h is the vertical distance between
the ball and the centrifugal fan.
* Gravity force: Fg9 wr3 Pbg, where Pb is the ball
density, and g is the gravity constant.
Therefore, the resulting vertical net force, FR (h), act-
ing on the ball can be computed as follows:
FR(h) Fg FVP+Fdrag
and by applying the Newton's Second Law, the vertical
motion of the ball can be modelled. It is worth pointing
out the dependency of the air flow velocity, v, on the dis-
tance to the fan output, h. Although there exists a wide
zone where the velocity can be considered to be propor-
tional to h-', if h is large enough (in this application,
about 50 cm) the flow velocity decreases drastically, be-
coming its dynamics very complex.
From the above equations, and taking into account that
other effects (such as lateral and rotational motions, lack
524
of sphericity, rugosity, etcetera) have not been considered,
it is easy to realize that having an accurate model of the
object motion results very complicated. Furthermore, the
flow velocity at the fan output is modified by means of the
variable-speed drive, but no measurement of the velocity
is available. This implies that the fan dynamics should be
also modelled.
Due to the complexity of the system, some linear mod-
els are derived at several operating points in order to syn-
thesize some model-based controller. Thereby, excited
enough control signals around three equidistant operating
points are supplied. From input-output data analysis, and
by using an appropriate software package for example,
[6], transfer functions are obtained, whose frequency re-
sponses are plotted in figure 4. A nominal model, whose
frequency response is also included in figure 4 (thick line),
is chosen as follows:
-0.4023s4 + 13.97s 3- 111.882 - 533.3s + 5881
GN(S) s4 + 45.82s3 + 178.3s2 + 998s + 2542
30ii
20. ,
1~~ ~
0
O 0
.
I , 4 -
Figure 4. Frequency responses at several
operating points
It can be checked that the system is stable, with a domi-
nant pair of complex poles, which makes the step response
of the system be very oscillatory (see figure 5).
3 Control strategies
The following three control strategies has been imple-
mented in this system: a classical PID controller, a robust
control law based on the Ho, mixed sensitivity approach
and a controller based on fuzzy logic.
Two kind of experiments are carried out in order to
test the performance supplied by these control strategies:
tracking experiments applying a step-change of 20 cm at
the position reference, and disturbance rejection experi-
ments, in which some air flow is removed by hand during
a small time interval.
VOLUME 1
 signals, the error vector, the measured variable and the
control signal respectively.
60
Output (cm)

50

40

30
Input (Hz)

20 K(s
IU.-
G(s

time (s)

Figure 5. Step of the nominal

Figure 7. SIKSIT mixed sensitivity conf igu-
model
response
ration

The design of appropriate weighting functions has been

3.1 PID control law
carried out by means of the procedure exposed in [7].
Next, steps followed to synthesize the H,,,, controller are
The first strategy implemented is based the classi-
exposed:
on

cal Proportional-Integral-Derivative controller. A first set

of values for its gains is performed by means of the well- 1. Scaling and estimation of the multiplicative
known Ziegler-Nichols tuning rules; afterwards a fine ad- output uncertainty
justment is fulfilled to attain a better behavior.

Figure 6 shows some experimental results obtained

At this first step, the identified transfer functions as
with this control strategies. In this case, it can be observed
well as the nominal model selected must be scaled.
a good tracking response, while the disturbance rejection This is carried out, as stated in [10], using the max-
experiment exhibits a slow and too oscillatory behavior.
imum desired variation of control signal, as well as
the maximum allowed variation of the output.
Using these scaled functions, the multiplicative un-
certainty with respect to the nominal model is esti-
mated as follows:
E0js (G6(s - NN(s))G6N(

where GN (s) represents the scaled nominal model

-.41
40 - and Q (s) stands for the p different scaled non-
-Fz
:-38 - nominal systems, one for each operating point,
LD)
-5
I36 -
where the controller is required to work properly.
.m
0
:34 -
0
32
---
3 0
0 5 10 15 20 25
2. Design of weighting functions
time (s)

Figure 6. Experimental results with the PID Once the multiplicative uncertainty has been esti-
controller mated, function WT (s) in such a way that it must be
stable, minimum phase, with high gain at high fre-
quencies, and with module greater than the module
of the uncertainty previously computed for each non-
3.2 H,,,, controller
nominal model and frequency, that is,
mixed
A robust H,,,, control

sensitivity problem [10]

law based on the

has been also tested. With

SIKSIT ~WT(j'w) . E0jj( w) Vw, Vp

this control strategy, closed-loop sensitivity functions

For this application, the following design is carried
are

shaped using some weighting functions. This configura-

out
figure 7, P(s) is the so-called ge-
(0.66s + 1)
tion is shown in where

neralized plant, and WT (s) =0.520.ls

w, z, v and u represent the exogenous
+1

525 ~~~~~~~VOLUME 1
 which fulfills the above constrains, as shown in figure are shown in figure 9. It can be seen some tracking im-
8. provements in relation to the ones of the PID controller,
while poor results are achieved in disturbance rejection
experiments. The time response becomes too oscillatory,
and with a great amplitude. This may be explained keep-
ing in mind that this is model-based controller, which tries
to cancel the system dynamics. In this way, it must be re-
membered that the system has a dominant and very oscil-
latory mode. These poles vary depending on the operating
point. Therefore, the oscillations may be caused by an im-
perfect cancellation of these poles.
80

60

v /,\,^,
40 - ,<\ /A-- v v ^,\ 'V
-\ ,a ,\^,n/
V VI AX,
kv tV
\\A
\ disturbance
\tracking /Irejection
Figure 8. Function WT(s) as an upper bound u ,1I
of the estimated multiplicative uncertainty 0 5 10
time (s)
15 20 2

42 /
I

Regarding to the weighting function Ws(s), its de- 408 -

sign is proposed as following expression: 236

U.
38 -
> I
g ~' "
c5
c~34
10o(I
-

WS(S) = as
+ )wT 32 -

+ 1310(r_-1)WT 30
0 5 ~~0 ~ ~~115 ~~10
~ 220 2E?5
time (s)

where WT 2.8 radjs is the crossover frequency of Figure 9. Experimental results with the Ho,
the transfer function WT(s). The following values controller
are proposed for the parameters: a 0. 5, 10-4,
and r, initially equal to 0. The final form of Ws (s)
is obtained by adjusting the value of r, under instruc-
tions of the following rule of thumb: as the value of 3.3 Fuzzy control
r,increases, a faster (although more oscillatory) res- The third control strategy implemented is based on
ponse is attained. fuzzy logic [9, 5]. In particular, a pseudo-linear direct
control has been designed, which is, in some sense, simi-
Finally, the weighting function WKS (s) is chosen as lar to a PI controller [4]. Figure 10 shows the structure of
the identity in order to avoid numerical problems [3].
a general fuzzy controller.

3. Buliding up the generalized plant and obtaining

the controller zt ~~Set of
lz~~~ rules

As a final step, the generalized plant P(s) must N

be built up using the nominal model and weighting
functions previously designed (see figure 7). Taking
as starting data the generalized plant, the controller
may be attained by means of some synthesis method
implemented in some software package, such as [3],
and [2].
At this point it is important to remember that the
controller has been computed by means of a scaled
model. Thereby, an inverse-scale process must be
performed on the controller before implementing it
in the real plant.
The tuning procedure for the parameter r, has been per-
formed by direct inspection of the experimental response.
N Inference
41 system a
Figure 10. Structure of a fuzzy controller
In the case of a direct controller, the inputs only depend
on the error, while the output is oriented towards the ma-
nipulated variable. In this work, an incremental controller
has been synthesized, whose inputs are the error and its
derivative, weighted by the parameters K1 and K2 respec-
tively (see figure 11). The output is the derivative of the
manipulated variable, which is weighted by the parameter
K3.
On the one hand, two triangular membership functions
for each input are established, with their peak values at -

Experimental results obtained with this control strategy 1 and 1 respectively. On the other hand, three singletons

526 VOLUME 1
 e response is similar to the ones attained by the previous
controllers, but the amplitude of the oscillations decreases
FIS 1/s
a lot, at least in disturbance rejection experiments.
d/dtP KIntegrator

Derivator

Figure 11. Structure of an incremental fuzzy 60 -

controller 'o
4040- ' * X-

20 / \-disturbance
tracking rejection
o0 / I \I20
are employed for each output, whose values are -1, 0, and 5 10
time (s)
15 2

1 respectively. The following names are adopted for the

42 ~ /I l \
inputs: positive and negative; and the following for the 40 -
outputs: negative, zero, and positive. I~
38 -

wF
34 -

32 -
Negative Positive
30
0 5 10 15 20 2 ?5
time (s)

-1 0 1
Figure 13. Experimental results with the
Input fuzzy controller

Negative
___________j- Zero Positive
3.4 Comparison of results
In order to compare the results attained by the control
-1 0 1 strategies, the ITAE index of the time responses are ex-
Output posed in table 1.

Figure 12. Input and output membership Tracking Disturbance

functions
The base for this controller consists of the following
four rules:
* If e is positive and de is positive then du is positive
* If e is positive and de is negative then du is zero
* If e is negative and de is positive then du is zero
* If e is negative and de is negative then du is negative
Despite of the resultant fuzzy controller is nonlinear,
there are some similarities among the parameters of these
controller and the ones of a PID. In fact, once the parame-
ter K1 has been adjusted, the parameter K2 plays the role
of the integral time, whilst the parameter K3 is similar to
the proportional gain.
The following rule is employed the defuzzification pro-
cess:
n avoid the damage of the system.
Ei=1 Hi
where n is the number of singletons (in this case, three:
-1, 0, and 1) and ,ui is the membership grade associated to
the i singleton.
Experimental results attained with this control strategy
are shown in figure 13. In this case, the speed of the time
rejection
PID 19.42 10.88
Hoc 17.54 16.12
Fuzzy 18.66 9.72
Table 1. ITAE index for each controller
It can been observed that the implemented PID con-
troller supplies a good trade-off between tracking and dis-
turbance rejection problems. The Ho, controller is able
to improve the tacking performance, but at the expense
of a poor performance in the disturbance rejection. Fi-
nally, it can been noted that the implemented fuzzy con-
trol law slightly improves the performance supplied by the
PID controller in both cases.
It is worth pointing out that both PID and fuzzy con-
troller provide smooth control signals (see figures 6 and
13 ). However, the control signal generated by the Ho,
controller has a significant high frequency component (see
figure 9). This fact should be kept in mind in order to
4 Conclusions
A lab plant, which consists of industrial components,
has been used to study suitable control strategies for the
position control of pneumatic levitation systems. A brief

527 VOLUME 1
modelling study has been performed, and linearized sys-
tems around several operating points has been obtained
from analyzing input-output data.
Three control strategies has been implemented, and ex-
perimental results for tracking and disturbance rejection
have been presented. The first controller has been a clas-
sical PID, which has provided a good trade-off between
tracking and disturbance rejection.
A pseudo-linear direct control based on fuzzy logic has
been also tested. Despite of few rules have been designed,
with triangular membership functions for the inputs and
three singletons for the output, this controller has slightly
improved the performance attained with the PID.
Finally, a model-based Ho, controller has been also
tried out. Although better behavior has been achieved for
tracking experiments, this controller has provided a poor
performance for disturbance rejection. This fact may be
explained taking into account the uncertainty of the sys-
tem dominant mode, which is too oscillatory.

Acknowledgments
The authors would like to acknowledge MCYT and
FEDER for funding this work under grants DPI2003-
00429 and DPI2004-06419. Likewise, the authors appre-
ciate the support of the DMASTEC organization.

References
[1] Algarin, D. Reconocimiento de objetos mediante visi6n
artificial (in Spanish). Proyecto Integrado, Centro Tec-
nol6gico ALTAIR, 2002. In Spanish.
[2] Balas, G., Doyle, J., Glover, K., Packard, A., Smith, R. ut-
Analisys and Synthesis Toolbox User's Guide. The Math-
Works Inc., Natick, Mass, second edition, 1995.
[3] Chiang, R., Safonov, M. Robust Control Toolbox User's
Guide. The MathWorks Inc., Natick, second edition, 1998.
[4] Jantzen, J. Tuning of Fuzzy PID Controllers. Technical
University of Denmark. Departament of Automation, Bldg
326, DK-2800 Lyngby., second edition, 1998. Tech. Re-
port no 98-H 871.
[5] Leondes, C.T. Fuzzy Theory Systems. Techniques and Ap-
plications. Academic Press, 1999.
[6] Ljung, L. System Identification Toolbox. The Math Works,
Inc., 1995.
[7] Ortega, M.G., Rubio, F.R. "Systematic Design of Weigh-
ting Matrices for H,, mixed Sensitivity Problem". Jour-
nal of Process Control, 14(1):89-98, 2004.
[8] Reed, J. Biscuit Levitation by airjets. Biennial Report
1998-2000, Silsoe Research Institute, New York, 2000.
[9] Rubio, F.R., Berenguel, M., Camacho, E.F. Fuzzy Logic
Control of a Solar Power Plant. IEEE Trans. on Fuzzy
Systems, 3(4), 1995.
[10] Skogestad, S., Postlethwaite, I. Multivariable Feedback
Control. Analysis and Design. Jonh Wiley & Sons, New
York, 1996.

528 VOLUME 1

View publication stats